Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics

Business Intelligence for Marketing

Business intelligence for marketing is the layer between raw platform exports and the decisions a team makes: a warehouse that keeps history, definitions that make each metric mean one thing, queries that join spend to outcomes, and dashboards that someone owns. This hub explains how the pieces fit, introduces the guides in this section, and helps a team decide whether it needs any of it yet.
SectionintermediateUpdated

Learn this as a path

Prefer a guided order? These learning paths read pages from this section in sequence, with exercises and checkpoints.

In this section

Guides · 10

Reporting and BI in one paragraph

A reporting setup reads each platform through its own interface or a connector: Google Ads describes what Google Ads did, GA4 describes what the website did, the CRM describes what sales did. Each is accurate about itself and silent about the others, and each keeps only as much history as it chooses to. Business intelligence copies the data out of every system into a store the team owns, keeps every day of it, joins spend to sessions to leads to revenue on shared keys, and publishes metrics whose definitions were agreed once and written down. The difference is not the chart at the end. It is whether the number behind the chart can be traced to its rows, joined to another system's rows and reproduced next month by a different person. The reporting versus BI comparison sets the two side by side; the reporting and BI overview covers the consumption side, meaning audiences, cadence and narrative. This section covers production: how trustworthy numbers get made.

The marketing BI stack

Every marketing BI setup, whatever the vendors, has the same layers. Teams differ in how many they build themselves and how many they buy, and in how far down the stack they have got.

LayerWhat it doesWhat marketing puts in itGuide
SourcesThe systems that generate dataAd platforms, GA4, CRM, e-commerce platform, email tool, call trackingReporting vs BI
IngestionCopies source data into the warehouse on a scheduleManaged connectors, the GA4 BigQuery export, platform APIs, spreadsheet uploads for the restBigQuery export basics
WarehouseStores raw and modelled tables with full historyDaily spend, events, sessions, orders, customers, campaign metadataSQL for marketers
TransformationCleans, standardises and joins raw tables into martsChannel mapping, UTM parsing, currency normalisation, spend-to-outcome joinsSQL for marketers
Metrics layerHolds each metric's definition once: formula, grain, filters, ownerCost per lead, return on ad spend, blended acquisition cost, pipeline by sourceMetrics layer and definitions
BI toolQueries the layer and renders dashboards and explorationsLooker Studio, or a licensed BI product once the team outgrows itLooker Studio section
ConsumptionDashboards, scheduled reports, ad hoc queries, alertsWeekly performance views, monthly reviews, finance reconciliationsDashboard governance

Most marketing teams enter the stack from the bottom up: Looker Studio reading directly from connectors, then the GA4 BigQuery export as the first warehouse table, then spend tables beside it, then definitions once two dashboards disagree. That order is fine; the mistake is treating any one stage as finished. The arithmetic the warehouse models is the arithmetic of the KPI framework: outcomes decomposed into drivers, which is what a KPI tree writes down before anyone builds a table.

The core guides

  1. Reporting vs business intelligence — what each is for, where reporting is enough, what BI adds, and how teams move from one to the other without rebuilding everything at once.
  2. SQL for marketers — the handful of query shapes that answer most marketing questions: filtering and aggregating, time grains, joining spend to outcomes, window functions for cohorts, and the pitfalls that inflate totals.
  3. Metrics layer and definitions — why the same metric differs by tool, how to write a definition with formula, grain, filters and owner, and how to version and enforce it.
  4. Dashboard governance — ownership, certification, an inventory, naming and access rules, change management and the review that retires what nobody uses.

Two neighbouring sections do work this one depends on. The Looker Studio section covers the dashboard tool most marketing teams use first, including data sources, blending and performance. The attribution section covers the modelling question that BI projects usually end up being asked to answer, because once spend and outcomes sit in one table, someone will ask which channel deserves the credit (Attribution). Both also have a tutorial that works on warehouse tables: moving a Looker Studio report onto a warehouse and building an attribution model in SQL from the Google Analytics BigQuery export.

Further guides

Further guides go deeper into building the tables, reading what they show, checking and correcting what is shared, and running the team that uses them:

  • Designing a marketing data model — the tables behind marketing reporting: facts for spend, sessions, leads and orders, shared dimensions for date, channel and campaign, and the keys that join them.
  • Loading ad platform data into a warehouse — connectors or exports, refresh windows for late conversions, currency, time zones and renamed campaigns.
  • Writing data tests for marketing tables — a tutorial for SQL tests that catch broken data before a dashboard does: duplicates, missing values, unmapped rows, freshness, volume and reconciliation.
  • Explaining a change in a marketing metric — rule out data changes, split the metric into parts, separate mix from rate, find the segment and test the story.
  • Comparing periods in marketing reports — weekdays and holidays aligned, partial periods, seasonality, tracking breaks and late conversions.
  • Reading small numbers in marketing reports — why rates built on small counts jump around, ranges instead of single figures, minimum volumes, pooling and control limits.
  • Running an analytics request process — intake that captures the decision, triage and sizing, self-serve first, reusable answers and closing the loop.
  • Building data literacy in a marketing team — the skills that matter, definitions within reach, reading uncertainty, office hours and checks that it worked.
  • Reviewing an analysis before it is shared — a tutorial for a second person's review of a one-off analysis: restate the question and the decision, reproduce the headline numbers, check definitions, joins and the counts behind every rate, pair each claim with its evidence and record the sign-off.
  • Correcting a number after a report has gone out — a tutorial for a wrong figure in a sent report or live dashboard: confirm and size the error, fix the source, restate the figures with the old ones still findable, tell the people who acted on it first and add the test that would have caught it.

When a team needs BI

The honest answer for many teams is: not yet. A single channel, one platform and a reviewer who reads the platform's own reports is a reporting problem, and a warehouse would add cost without adding decisions. The signals that the balance has tipped are specific:

  • The same metric differs between meetings. Two dashboards, two connectors, two definitions of a conversion. Nobody can say which is right because neither can be traced.
  • Questions need joins. Cost per lead by landing page, revenue by first-touch channel, retention by acquisition campaign. Each needs rows from two systems on a shared key.
  • History matters more than the platforms keep. Year-on-year comparisons, seasonality, cohorts that take a year to mature.
  • Several people build their own reports, each from their own export, and the reports drift apart.
  • Finance asks for reconciliation, and the marketing number has to tie to the invoice and the ledger.
  • Attribution and incrementality questions cannot be answered from any one platform's view of itself.

When several of these are true at once, the cost of not having BI is already being paid in meetings, re-exports and wrong decisions. When none are, the right move is better reporting discipline: one dashboard, one owner, one definition of each KPI.

Illustrative example — not a real result

A subscription box deciding whether to build BI

A subscription box business runs paid search, paid social and affiliate, with a CRM for subscribers and a spreadsheet for refunds. Monthly reporting takes the marketing lead two days: four exports, three pivot tables, one deck. In the latest review paid social shows 1,420 new subscribers at £38 each while finance counts 1,180 new paying subscribers across all channels, and nobody can reconcile the gap. The team costs the alternative: a managed connector for each ad platform, the GA4 export into BigQuery, a weekly SQL model that joins spend to subscriber records on the order id, and one Looker Studio dashboard reading from it. The build is estimated at six weeks of a part-time analyst and a monthly warehouse bill under £150. Against two days a month of the lead's time and a subscriber count that is wrong by 20 percent, the decision makes itself.

Common BI problems

Most BI problems show up as one of a few symptoms. The table maps each to the guide that diagnoses it.

SymptomUsual causeGuide
The same metric differs by dashboardNo shared definition; each tool computes its own formula, window or filterMetrics layer and definitions
Query totals exceed what the platforms reportJoin fan-out, double counting across grains, missing distinct countsSQL for marketers
Dashboards nobody trusts or ownsSprawl, copies of copies, no certification or retirementDashboard governance
The team cannot answer joined or historical questionsStill on platform reporting; no warehouse and no historyReporting vs BI
The raw export exists but nobody uses itNo SQL skill in the team; no modelled tables between export and dashboardSQL for marketers, BigQuery export basics
Dashboards are slow or time outBlending and calculating in the BI tool instead of modelling upstreamLooker Studio performance
Finance and marketing numbers do not tieDifferent grain, time zone or counting unit on each side, undocumentedMetrics layer and definitions
A Looker Studio chart disagrees with the GA4 interfaceDifferent dates, mixed scopes, thresholds, blends, averaged ratios or a stale cacheLooker Studio not matching GA4
Queries on the raw export disagree with the GA4 interfaceModelled data, session logic in SQL, time zones, source scope or incomplete tablesBigQuery export not matching GA4
Ad platform conversions in the warehouse fall short of the platform for recent daysConversions credited after the load, with no refresh window for late arrivalsLoading ad platform data
A broken table reaches a dashboard before anyone noticesNo tests on keys, required fields, mapping, freshness or volumeData tests for marketing tables
A dashboard keeps showing old figures, or its latest days never arriveA cache or extract, a stalled warehouse load, a fixed date range, provisional source days or an early deliveryLooker Studio report not updating
A metric moved and the first plausible story is taken as the answerNo method for ruling out data changes, separating mix from rate and finding the segmentExplaining a metric change
A comparison with last month or last year points the wrong wayMisaligned weekdays and holidays, partial periods, tracking breaks or seasonalityComparing periods
Campaigns paused or rewarded on a handful of conversionsRates built on small counts read as newsReading small numbers
The analytics backlog fills with charts nobody acts onRequests answered in arrival order, without the decision behind themAnalytics request process
An analysis reaches a decision maker checked only by the person who wrote itNo second-person review of definitions, joins, comparisons, the counts behind rates and causal wordingReviewing an analysis
A wrong number is found after the report has gone outA double-counting join, a dropped filter or an early export, fixed quietly instead of restatedCorrecting a number

For the consumption side of the same work, meaning the audiences a dashboard serves, the cadence it runs on and the narrative that goes with the numbers, the reporting and BI overview is the companion page to this section.

Terms used on this page

Frameworks

Related